Mangrove blue carbon in arid environments: soil carbon variability and predictive modelling
Abstract
The study explores spatial and depth-dependent variations in soil organic carbon (SOC) and soil physical properties in the hyper-arid mangrove ecosystem along the north-eastern United Arab Emirates (UAE) coastline. Results indicate that SOC was highest in inner plots, following the pattern Inner > Water > Land, with greater SOC in the 0–20 cm (topsoil layers) compared to 20–40 cm (bottom layer). The research also investigated the relationship between SOC and remotely sensed indices using Sentinel-2 and Landsat 8 imagery. Among the tested indices, the Green Normalized Difference Vegetation Index (GNDVI, ρ = 0.65) and Normalized Difference Water Index (NDWI, ρ = -0.65) exhibited the strongest correlations with SOC. To model SOC variability, three advanced machine learning algorithms, XGBoost, Random Forest, and Gradient Boosting, were applied. XGBoost achieved the highest predictive performance (R² = 0.711), outperforming Random Forest (R² = 0.684) and Gradient Boosting (R² = 0.680). These results demonstrate the potential of integrating remote sensing and machine learning for accurate SOC estimation in hyper-arid mangrove ecosystems. The findings underscore the critical role of mangroves in blue carbon storage and highlight the utility of remote sensing technologies for long-term SOC monitoring, conservation planning, and climate change mitigation.